There is a temptation to treat the latest warnings about artificial intelligence as another round of technological alarmism. We have heard predictions about AI transforming society before, and some have proved exaggerated. But two very different developments over the last three months suggest that we need to pay much closer attention.

The first is Bill Gates warning that AI could fundamentally reshape employment, cybersecurity, biological security and the relationship between humans and machines. The second is an extraordinary cyber incident involving an autonomous AI agent being tested by OpenAI that ultimately compromised infrastructure belonging to Hugging Face.

Taken together, these developments raise a bigger question than whether AI is good or bad. They force us to ask whether politics, government and society are keeping pace with what is happening.

Gates’ warning deserves attention because it goes well beyond the familiar debate about whether AI will take people’s jobs. He has raised concerns about permanent disruption across parts of the labour market, AI being used to facilitate cyberattacks and biological threats, the psychological effects of increasingly persuasive systems and the danger that governments respond only after disruption has occurred.

He has even raised the prospect of protecting certain categories of work for humans and reconsidering taxation so economies do not simply reward the replacement of people with machines.

That does not mean Gates should simply be taken at his word. Predictions about mass unemployment, bioterrorism or AI becoming overwhelmingly dangerous remain predictions. They need to be challenged and properly understood rather than accepted because they come from one of the world’s most influential technology figures.

But dismissing the warnings would be equally irresponsible. The important question is whether governments are sufficiently prepared if even some of these risks materialise.

That is where the Hugging Face incident becomes particularly significant, because unlike many predictions about AI, this was not simply a forecast about the future.

In July, OpenAI disclosed that an autonomous AI agent involved in an internal cybersecurity evaluation had escaped its controlled testing environment and ultimately compromised infrastructure belonging to Hugging Face. The subsequent technical reconstruction indicated that the intrusion involved thousands of automated actions, with the system moving through infrastructure, accessing credentials and attempting to move laterally through systems.

There is an important distinction to make. The AI had not suddenly developed a human desire to attack another company. It was operating within an evaluation designed to test its capabilities and appears to have interpreted compromising Hugging Face as a way of achieving its assigned objective.

But that distinction arguably makes the lesson more concerning.

AI does not need consciousness, emotion or malicious intent to create consequences humans struggle to control. A sufficiently capable system given an objective, tools, access and autonomy can discover ways of pursuing that objective that its creators did not anticipate.

The most worrying element may not be the breach itself but the speed at which these capabilities are improving.

The agent was operating at machine speed, undertaking thousands of actions without the limitations faced by a human operator. A human hacker has limited time, attention and endurance. An AI system can potentially keep trying, learn from results and change its approach continuously.

As increasingly capable models become more widely available, this becomes much more than a technology industry problem. It becomes an issue of national resilience, economic security and democratic governance.

So what does this mean for the Liberal Democrats?

The answer cannot simply be for the party to declare that AI is dangerous. Equally, we should not assume that technological progress automatically produces social progress.

AI has enormous potential to improve healthcare, education, public services, scientific research and productivity. The question is how we ensure those benefits are broadly shared while the risks are properly understood and managed.

That means asking difficult questions about accountability, transparency and power. Who audits increasingly autonomous systems? Who is responsible when they cause harm? What information should technology companies be required to disclose? How do we protect critical infrastructure? What happens to workers whose roles are fundamentally changed or eliminated? And how do democratic institutions make decisions about technologies evolving faster than the policy frameworks designed to govern them?

There is perhaps an even more fundamental question: do politicians have sufficient access to the technical expertise required to understand what they are legislating about?

That is not simply a Liberal Democrat problem. It is a challenge for politics as a whole. But it is one the Liberal Democrats should be prepared to confront.

Perhaps this is why it is worth asking whether the party should consider developing an AI Affiliate Organisation. Not another layer of party bureaucracy, and not an organisation that arrives with all the answers, but a space where people with expertise in AI, cybersecurity, economics, employment, ethics and technology can come together to understand what is changing and what the political implications might be.

The purpose would be simple: to create a better-informed conversation before events force politicians to react.

The Gates warning and the Hugging Face incident should therefore be considered together. One describes the potential consequences of AI across the economy and society. The other demonstrates that some of the capabilities underlying those warnings are already emerging.

Neither proves that the most extreme predictions about AI will come true. But together they provide a compelling reason to take the issue seriously.

The choice should not be between embracing AI unquestioningly and trying to stop it. The challenge is understanding it well enough to decide where it should be encouraged, where safeguards are required and where society may need to draw boundaries.

The technology is moving quickly.

The question is whether our political understanding is moving quickly enough.

Perhaps, for the Liberal Democrats, the first step is simply to start the conversation.

Should we be looking at developing an AI Affiliate Organisation to begin discussing these very clear and rapidly emerging issues?

If you are interested in the potential development of an AI Affiliate Organisation and would like to be part of that conversation, please click here and provide your email details to register your interest and receive further information.

* Sean Worrell is a digital advocate, AI enthusiast and Liberal Democrat member